For most of the last few years, using AI meant going somewhere private. You opened a tab or an app, typed a prompt, read the reply and carried whatever you learned back to your team yourself. The AI was a tool you used alone, and the value it produced depended entirely on one person remembering to go and ask. That pattern is now breaking, and Anthropic's Claude Tag is the clearest signal yet of what replaces it.
Claude Tag drops a full Claude Code teammate directly into a Slack workspace. Rather than being another window to prompt, it behaves as a persistent, asynchronous entity that picks up context across channels, breaks a tagged request into stages, and does the work in the thread, whether that is writing a pull request, running an analysis, or following up on a quiet conversation. Anthropic says its own internal version now generates roughly 65 percent of its product team's code, and one team member describes doing more than 90 percent of his work through it. The AI researcher Andrej Karpathy framed it as the third major redesign of how we interact with language models, after the website you visit and the app you download.
This is not a far-off idea for us, it is a road we are already travelling. Aphelion AI is a private enterprise AI platform built to deploy your AI agent inside infrastructure you own and control, and group chats that include AI agents are already part of what we offer. The shift Claude Tag points toward, from a private chatbot to a shared teammate working in the open alongside a team, is one we have been building deliberately rather than discovering by surprise.
From a Private Chatbot to a Shared Teammate
The most useful way to read Claude Tag is not as a single feature but as a set of shifts in how people relate to AI at work. The commentary around its launch identified five worth naming plainly:
- From app-native interfaces to existing workplaces: the AI comes to the group chat your team already lives in, rather than asking everyone to learn a new destination.
- From a private chatbot to a shared teammate: the agent is something the whole team can see, tag and build on, not a personal assistant hidden in one person's browser.
- From single-user context to full-team context: the agent reads the surrounding conversation, so it understands what the group is trying to do without being briefed from scratch each time.
- From prompting to delegation: people increasingly tell the agent what they are trying to accomplish and let it figure out the steps, rather than dictating each instruction.
- From a personal tool to an organisational dependency: once a team relies on the agent for status updates, fixes and follow-ups, it stops being a nicety and becomes part of how the organisation runs.
Every one of those shifts is something a group chat with an embedded agent makes possible, and every one of them is already live in the Aphelion model. When an agent sits inside the conversation, the context is shared by default, delegation feels natural because the whole team can watch the work happen, and the agent becomes a genuine participant rather than a private utility.
The headline is not that AI can now write code from a chat message. It is that AI has moved out of a private window and into the shared space where teams actually coordinate. That change in location changes everything about how the work gets distributed, watched and trusted.
Why the Group Chat Is the Right Place
There is a reason this redesign landed where teams already talk. A coworker who only ever speaks to you one to one, in a room nobody else can enter, is hard to delegate to and impossible to supervise. Move that same coworker into the room where the team plans, debates and tracks its work, and suddenly delegation, visibility and accountability come for free. The observers describing Claude Tag kept reaching for the same word: it behaves like a colleague that picks things up.
That is exactly the design principle behind Aphelion's group-chat agents. An agent invited into a team conversation can take an objective, break it into stages and carry out multi-step work where everyone can follow along, the same way a human teammate would. Because it shares the channel's context, it does not need to be re-briefed for every task, and because the work happens in the open, the team can correct course early rather than discovering a problem after the fact. The result is the delegation model that Claude Tag gestures at, delivered as a standard part of the platform rather than a bolt-on.
It also matters that the agent can act, not just talk. A shared teammate that can only answer questions is a search box with manners. The value arrives when it can reach the systems a team depends on, which is why Aphelion's approach to data enrichment and private model hosting underpins the in-chat experience. The agent draws on the context and records your business already holds, so its contributions in the conversation are grounded in your real data rather than generic guesswork.
The Question Nobody Should Skip Over
A shared agent has an uncomfortable property that the launch commentary did not shy away from: it reads everyone's messages. One sceptic put it sharply, warning that whoever introduces the agent risks becoming the person who brought a surveillance device to the meeting. When the channel in question runs on an external service, that concern is not paranoia, it is a legitimate governance question about where sensitive conversation ends up.
This is the point where ownership stops being a philosophical preference and becomes a practical safeguard. The hidden costs of a shared agent on someone else's platform are worth naming:
- Data exposure: every prompt, document and message the agent reads passes through infrastructure you do not control, which complicates GDPR, HIPAA and ISO obligations.
- Vendor lock-in: once a team depends on a hosted agent, the provider can change pricing or terms on a capability your operations now rely on.
- Usage-based cost: a metered agent grows more expensive precisely as it becomes more useful, taxing your own success.
- Audit friction: proving compliance means vetting a third party's systems you cannot see, rather than reviewing your own logs.
Aphelion removes each of these by keeping the agent and the conversation inside your governed environment. Because the model runs on compute you own or exclusively control, the shared context never leaves your walls, audits become a review of your own controls, and the flat per-user fee means the bill does not climb every time the agent proves its worth. Where integration is needed, Aphelion treats system integration as a core platform capability, so connecting the in-chat agent to a CRM, ERP or document store is incremental work rather than an expensive special project.
"A shared AI teammate is only as trustworthy as the place it lives. Put it inside infrastructure you own, and the surveillance question answers itself, the data stays governed and the capability stays yours."
Owned, Not Rented, by Design
One voice in the launch coverage made the broader case directly: in a market where access to a hosted model can be suspended or repriced, there are increasingly good reasons not to outsource your shared agent to a single closed provider. Build or own it instead, and you get the model you want, full customisation, no lock-in and no waitlist. That argument is the foundation Aphelion was built on, well before group-chat agents became the headline.
The contrast is straightforward. A hosted feature like Claude Tag delivers the shared-teammate experience while keeping the model, the context and the pricing power on a provider's infrastructure. A private Aphelion agent delivers the same in-chat experience while keeping the model on compute you control, the data inside your environment and the cost on a flat per-user fee. For a casual, low-stakes use case the hosted route is perfectly reasonable. For a team handling sensitive work or expecting sustained volume, the owned approach keeps cost predictable and data private while delivering the identical shift in how people work.
The Same Shift, on Your Terms
To make the comparison concrete, here is how the two approaches line up on the dimensions that decide whether a shared agent is safe to depend on. Aphelion is charged at a flat per-user fee covering unlimited projects, chats and agents, with data enrichment and integrations scoped to each customer's needs.
| What matters | Aphelion private agent | Hosted in-chat agent |
|---|---|---|
| Where the conversation lives | Your own infrastructure | Third-party servers |
| Shared, in-chat teammate | Included as standard | Yes, provider-hosted |
| Pricing as usage grows | Flat, headcount only | Scales with usage |
| Unlimited projects, chats and agents | Included | Often metered or tiered |
| Compliance and audit posture | Review your own controls | Vet a third party |
| Vendor lock-in risk | None, capability is yours | Dependent on the provider |
| System integration | Core platform capability | Bundled or bespoke, variable |
The point is not that the shared-teammate model is wrong. It is that the model is clearly the direction of travel, and the only real question is who controls the infrastructure it runs on. Aphelion's answer is that it should be you.
Frequently Asked Questions
What is Claude Tag and what does it change about using AI?
Claude Tag puts a full Claude Code teammate inside a Slack workspace, so instead of opening a separate app to prompt an assistant, you tag the AI in a group chat and it picks up the surrounding context, breaks a request into stages, and does the work in the thread. The wider significance is a shift in how teams interact with AI, moving from a private chatbot used by one person to a shared, persistent entity that works alongside everyone, and from prompting step by step toward delegating an outcome. Aphelion is already on this path, with AI agents that operate inside group chats as part of the standard offering.
How does Aphelion AI deliver AI agents inside group chats?
Aphelion provides private AI agents that live inside team group conversations rather than in a separate window, so a colleague can bring the agent into a discussion, give it an objective and let it carry out multi-step work where the team can see it. Because the agent runs on infrastructure you own or exclusively control, the shared context never leaves your environment, and the same flat per-user pricing covers unlimited projects, chats and agents. This delivers the shared-teammate model that Claude Tag points toward, without handing your team's conversations to a third-party platform. You can read more on the AI Agents page.
Is a shared AI teammate in group chat safe for compliance and sensitive data?
A shared agent reads everyone's messages in a channel, which raises real privacy and governance questions when that channel runs on an external service. Aphelion answers this by keeping the agent and every prompt, document and output inside your governed environment, so sensitive discussion is never routed to a shared external platform. That makes GDPR, HIPAA and ISO work a matter of reviewing your own logs and controls rather than vetting infrastructure you cannot inspect, which lowers both the cost and the risk of the compliance process.
Can a group-chat AI agent connect to the business systems a team already uses?
Yes. The value of a shared teammate comes from acting on real systems, not just answering questions, so it needs to reach the CRM, the document store and the other tools a team runs day to day. Aphelion treats system integration and data enrichment as core platform capabilities rather than bespoke add-ons, so the in-chat agent can pull context from and act on the systems you already use. Building on a modular platform means new connections are incremental work rather than expensive refactoring.
Claude Tag vs a private Aphelion agent: which should a team choose?
Claude Tag is a hosted feature tied to a single closed provider, which means your shared context lives on someone else's infrastructure and your pricing scales with usage. A private Aphelion agent gives you the same in-chat, shared-teammate experience while keeping the model on compute you control, removing per-token metering and vendor lock-in, and leaving the capability as an asset you own. For any team that handles sensitive work or expects sustained volume, the owned approach keeps cost predictable and data private while delivering the same shift in how people work with AI. You can learn more about the team behind the platform on our About page.
Where This Goes Next
Whether or not Claude Tag itself becomes the dominant product, the pattern it represents is not going away. AI is moving out of the private window and into the shared space where teams coordinate, and the labs and agent companies are all heading in the same direction. The interesting decision for any organisation is no longer whether to adopt a shared AI teammate, but on whose infrastructure it should run.
Aphelion's bet is that the answer is yours. Group chats with embedded AI agents are already part of how the platform works, delivered with the data privacy, predictable cost and ownership that a teammate reading your every message demands. The shift is real and it is happening now. The only question worth asking is who holds the keys to the room it works in.